|Publication number||US7639252 B2|
|Application number||US 11/161,669|
|Publication date||Dec 29, 2009|
|Filing date||Aug 11, 2005|
|Priority date||Aug 11, 2004|
|Also published as||US20060050072, US20100053158|
|Publication number||11161669, 161669, US 7639252 B2, US 7639252B2, US-B2-7639252, US7639252 B2, US7639252B2|
|Original Assignee||Ati Technologies Ulc|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (24), Non-Patent Citations (1), Referenced by (2), Classifications (5), Legal Events (3)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application claims the benefit of U.S. Provisional Application Ser. No. 60/600,940 filed Aug. 11, 2004, entitled “UNIFIED TESSELLATION CIRCUIT AND METHOD THEREFOR”, having as an inventor Vineet Goel, and owned by instant assignee and hereby incorporated by reference in its entirety.
The invention relates generally to apparatus and methods that perform tessellation on primitives such as those used in graphics processing systems, and more particularly to apparatus and methods that perform multiple types of tessellation.
Tessellation circuits and methods are known in graphics processing systems to divide primitives into subprimitives represented by subvertices. Generally, primitive tessellation can be done in three ways; these are known as discrete tessellation, continuous tessellation, and adaptive tessellation. As known in the art, tessellation may be carried out to break down a primitive, such as a quad primitive which is defined by four vertices, triprimitives which are defined by three vertices, a line which may be defined by two vertices, or any other suitable primitive, into smaller subprimitives in an effort to get a more accurate depiction of an outer shape of a primitive, displacement mapping, or for any other suitable purpose. For discrete tessellation, an integer tessellation level per primitive is used. For example, if a line is a primitive, and the tessellation level is one, the line is divided into two equally spaced segments. When the tessellation level is two, the line is broken into three segments. The level is taken from the perspective of an edge of a primitive. Similarly, a triangle is subdivided into four sub-triangles for level one tessellation.
For continuous tessellation, floating point or fractional levels of tessellation per primitive are provided so that for example a tessellation level of 1.1 or 1.2 may be used to get a finer granularity of a breakdown of a primitive into smaller subprimitives. Adaptive tessellation also employs a fractional level of tessellation for each edge of a primitive and typically includes the application of continuous tessellation to an inner portion of the primitive and a seaming tessellation at the edges of primitives. With adaptive tessellation, the process typically includes performing continuous tessellation by tessellating uniformly, a primitive shape within the primitive being tessellated and applying an adaptive tessellation technique to an area outside the uniformly tessellated area.
Known tessellation engines are typically limited in their operation. For example, a typical tessellation engine may only accommodate one type of primitive such as a triangle primitive. In addition, tessellation engines may also typically carry out only one type of tessellation such as discrete or continuous tessellation. However, software based adaptive tessellation techniques are also known wherein a host processor may carry out an adaptive tessellation operation per primitive. This is typically performed since implementation of a tessellation engine in hardware can require large amounts of memory but memory is typically limited in graphics processors or other devices that carry out some type of tessellation. The software adaptive tessellation approach may also carry out another type of tessellation namely continuous tessellation. However such known techniques typically use a forward difference method which can result in holes between shared edges due to error that can be generated utilizing the forward difference method. This can result in discontinuous image generations. The algorithms to avoid holes between adjacent primitive edges tend to be complex and costly if implemented in hardware.
Since a host processor such as a CPU is required to carry out the software based adaptive tessellation using the forward difference method, lower precision can typically result and also decrease the CPU performance since it is required to carry out the tessellation operation to determine the tessellation vertices which may then be passed, for example, to a graphics processor.
Accordingly, a need exists for a tessellation circuit that does not require a host processor to carry out tessellation vertex generation but that would also accommodate at least all three types of tessellation. Alternatively it would be desirable to provide a tessellation circuit that provided both continuous and adaptive tessellation but that did not unnecessarily burden a host processor in the computations, while at the same time, minimizing the hardware cost.
The invention will be more readily understood in view of the following description when accompanied by the below figures and wherein like reference numerals represent like elements:
Briefly, a hardware tessellation circuit serves as a unified hardware parametric coordinate generator for providing parametric coordinates for tessellation. The tessellation circuit includes control logic that receives tessellation instruction information, such as an instruction indicating which type of tessellation to perform on an incoming primitive wherein the different types of tessellation include discrete tessellation, continuous tessellation, and adaptive tessellation. The tessellation circuit also includes shared tessellation logic that is controlled by the control logic and includes a plurality of shared logic units, such as arithmetic logic units, that are controllable by the control logic based on the type of tessellation detected to be used for the incoming primitive. The shared tessellation logic is controlled to reuse at least some of the logic units for two different tessellation operations defined by the tessellation type information. The tessellation circuit can tessellate multiple different primitive types, such as lines, triangles, quad primitives, line patches, tri patches and rectangular patches. The tessellation circuit also generates associated parametric coordinates such as subprimitive coordinates in parametric space, for different types of tessellation wherein the different types of tessellation include, in one example, discrete tessellation, continuous tessellation, and adaptive tessellation. The generated subprimitive coordinates are then passed through a vertex shader or other suitable processing logic which computes the tessellation vertices for a desired patch or primitive.
In one embodiment, the shared tessellation logic includes set-up logic, edge walker logic, and output logic, wherein hardware in the set-up logic, such as ALUs, and hardware in the output logic, such as ALUs, are re-used for multiple different tessellation types. In addition, the state machine in the edge walker logic may share similar states for multiple different tessellation types. Accordingly, a more robust hardware tessellation engine is provided that operates on multiple primitive types and provides different tessellation operations while reusing some hardware to reduce costs and provide efficient operation. Other advantages will be recognized of ordinary skill in the art.
The tessellation circuit 10 also receives tessellation level data 18 which may be user selected through a suitable software driver or programmed into a register (not shown) which indicates the level of tessellation desired. For example, if the tessellation type is discrete tessellation, the tessellation level could be an integer such as 1, 2, 3, 4, etc. If the tessellation type data 14 indicates that continuous tessellation is to be applied to the given primitive, the tessellation level may indicate a floating point tessellation level per primitive, such as 1.2, 1.4, 3.2, or any other suitable floating point value. For adaptive tessellation one tessellation level per edge is sent to the tessellation engine.
The tessellation circuit 10 produces subprimitive parametric coordinates 20 for use by a tessellation vertex determinator. Depending upon the type of primitive, the parametric coordinates may be barycentric coordinates (UVW), UV coordinates or Ucoordinates or any other suitable parametric coordinates for the subprimitives. If desired, the tessellation circuit 10 may also produce reordered indices 22 that may be reordered for avoiding holes between edges of adjacent primitives. When the primitive type is a patch primitive (line, tri or rect patches), the tessellation circuit may also output quad identification data 24, which is used in reordering indices of patches in the evaluation shader.
The subprimitive coordinates 20 are produced in parametric space and are passed to, in one example, shader pipes, so the shader pipes can determine subtriangle vertice information in XYZ space. As such, the quad identification data 24 is a subquad identifier, which identifies a subquad primitive in the larger primitive identified by indices data 16. The quad id is used reordering indices of patches in order to avoid holes between two adjacent patches. The tessellation circuit 10 is implemented in hardware to effect a unified hardware parametric coordinate generator.
The control logic 200 may be any suitable logic that performs the functions described herein. Shared tessellation logic 204 is controlled by the control data 202 from the control logic 200 based on the type of tessellation detected to be used for the incoming primitive. The shared tessellation logic 204 reuses at least some logic units for multiple different tessellation types. For example, the shared tessellation logic 204 performs both continuous tessellation and adaptive tessellation operations and as such shares some of the same logic units for both types of tessellation.
The set-up logic 400 may be any suitable logic and computes Delta_U and Delta_V data based on one or more lookup tables. Delta_U and Delta_V are segments of a primitive in the UV coordinate space. The set-up logic 400 also computes the number of segments (NUM_Segments) in the UV directions for a given edge of a primitive. In addition, the set-up logic 400 computes the number of segments for an inner quad primitive for adaptive tessellation. Sections of the set-up logic are reused for all tessellation types and primitive types. The hardware reused is look up table, computing delta_u and delta_v and num_segments. An example of this is shown in the attached diagram (i), (ii) and (iii) in the TE spec.
The set-up logic 400 provides the number of segments data 406 to the walker logic 402 along with primitive type and tessellation type.
The walker logic 402 as shown in block 408, includes logic that determines the number of quads in a given primitive type. For example, if a triangle is the primitive type the walker logic determines the number of quads in such primitive being 3. The setup logic 400 also computes the number of segments for an inner quad, for example, where adaptive tessellation is required.
For example, referring to
Referring back to
The logic that carries out block 408 may be shared and reused for all tessellation types such as the states that control the determination of the quads for a given primitive type or compute the number of segments for an inner quad. As shown in block 414, if the tessellation type data indicates a continuation tessellation is to be performed, the walker logic processes inner quads by representing them in three states and sets V, for the UV equal to zero for line primitives. The resulting outputs are the UV reordered indices and any quad identifiers for patch primitive types.
Referring to block 416, if adaptive tessellation has been indicated by the tessellation type data 14, the walker logic processes horizontal strips of the primitive and as shown in block 418 processes vertical strips of the primitive that are outside the inner quad. For example, as shown in
The output logic 404 multiplies the UV coordinates from walker unit with the Delta_U and Delta_V values to compute actual u,v, parametric coordinate within a primitive for a sub-vertex. If the triangle or tri patch primitive type is used, u,v coordinates are converted to (r, s, t) coordinates which represent bary-centric coordinates for a sub-vertex within a triangular primitive. It also discards the V data for line primitive types and arranges output in a format acceptable by later processing such as a shader pipe which determines the vertices of the subprimitives identified by the parametric coordinates.
As such the setup logic and output logic each include logic units such as index reordering logic, barycentric coordinate generation logic, fix to floating point converting logic or any other desired logic that is controllable by control logic based on the type of tessellation detected to be used for the incoming primitive so that logic is reused.
The setup logic includes interpolators for computing delta_u and delta_v, using values from the lookup table. Such interpolators are used for all the primitive types and tessellation types. The shared tessellation logic produces barycentric coordinate data or reordered vertex indice data and quad primitive identification data depending upon the type of incoming primitive. The edge walker logic includes a state machine that carries out the operations shown in the flow chart in
The above detailed description of the invention and the examples described therein have been presented for the purposes of illustration and description only and not by limitation. Also, the word “coupled” means directly or indirectly coupled to facilitate operation of the circuit. It is therefore contemplated that the present invention cover any and all modifications, variations or equivalents that fall within the spirit and scope of the basic underlying principles disclosed above and claimed herein.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5357599 *||Jul 30, 1992||Oct 18, 1994||International Business Machines Corporation||Method and apparatus for rendering polygons|
|US5821940||Aug 3, 1992||Oct 13, 1998||Ball Corporation||Computer graphics vertex index cache system for polygons|
|US5914722 *||Apr 14, 1997||Jun 22, 1999||Ati Technologies Inc.||Memory efficient method for triangle rasterization|
|US6057848 *||Aug 27, 1997||May 2, 2000||Lsi Logic Corporation||System for rendering high order rational surface patches|
|US6078331||Sep 30, 1996||Jun 20, 2000||Silicon Graphics, Inc.||Method and system for efficiently drawing subdivision surfaces for 3D graphics|
|US6100894 *||Aug 27, 1997||Aug 8, 2000||Lsi Logic Corporation||Patch-division unit for high-order surface patch rendering systems|
|US6211883 *||Aug 27, 1997||Apr 3, 2001||Lsi Logic Corporation||Patch-flatness test unit for high order rational surface patch rendering systems|
|US6504537 *||Sep 5, 2000||Jan 7, 2003||Nvidia Corporation||System, method and article of manufacture for fractional tessellation during graphics processing|
|US6597356 *||Nov 21, 2000||Jul 22, 2003||Nvidia Corporation||Integrated tessellator in a graphics processing unit|
|US6600488 *||Nov 4, 2002||Jul 29, 2003||Nvidia Corporation||Tessellation system, method and computer program product with interior and surrounding meshes|
|US6624811 *||Aug 31, 2000||Sep 23, 2003||Nvidia Corporation||System, method and article of manufacture for decomposing surfaces using guard curves and reversed stitching|
|US6674433 *||Mar 10, 2000||Jan 6, 2004||Intel Corporation||Adaptively subdividing a subdivision surface|
|US6697063 *||Nov 25, 1997||Feb 24, 2004||Nvidia U.S. Investment Company||Rendering pipeline|
|US6819325||Mar 6, 2001||Nov 16, 2004||Microsoft Corporation||API communications for vertex and pixel shaders|
|US6906716 *||Apr 17, 2003||Jun 14, 2005||Nvidia Corporation||Integrated tessellator in a graphics processing unit|
|US6940515 *||Dec 3, 2003||Sep 6, 2005||Nvidia Corporation||User programmable primitive engine|
|US6965908||Mar 15, 2001||Nov 15, 2005||Micron Technology, Inc.||Multi-function floating point arithmetic pipeline|
|US7109987||Mar 2, 2004||Sep 19, 2006||Ati Technologies Inc.||Method and apparatus for dual pass adaptive tessellation|
|US7200532 *||Jun 16, 2003||Apr 3, 2007||University Of Kentucky Research Foundation||Subdivision surface-based geometric modeling system|
|US20040085313 *||Apr 17, 2003||May 6, 2004||Moreton Henry P.||Integrated tessellator in a graphics processing unit|
|US20040113909 *||May 9, 2003||Jun 17, 2004||Simon Fenney||Interface and method of interfacing between a parametric modelling unit and a polygon based rendering system|
|US20040227755 *||Dec 9, 2003||Nov 18, 2004||Adrian Sfarti||Bicubic surface real-time tesselation unit|
|US20060238535||Jul 5, 2006||Oct 26, 2006||Ati Technologies Inc.||Method and Apparatus for Dual Pass Adaptive Tessellation|
|EP0930582A2||Jan 19, 1999||Jul 21, 1999||Cagent Technologies, Inc.||A method and apparatus for arranging graphical data|
|1||Moreton, Henry; Watertight Tessellation Using Forward Differencing; SIGGRAPH Eurographics Workshop; 2001; pp. 25-32; NVIDIA Corporation.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US8482560 *||Apr 29, 2009||Jul 9, 2013||Intel Corporation||Image forming techniques|
|US20100164955 *||Apr 29, 2009||Jul 1, 2010||Sathe Rahul P||Image forming techniques|
|International Classification||G06T15/30, G06T15/00|
|Nov 16, 2005||AS||Assignment|
Owner name: ATI TECHNOLOGIES INC., CANADA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:GOEL, VINEET;REEL/FRAME:016786/0489
Effective date: 20050816
|Jun 25, 2009||AS||Assignment|
Owner name: ATI TECHNOLOGIES ULC, CANADA
Free format text: CHANGE OF NAME;ASSIGNOR:ATI TECHNOLOGIES INC.;REEL/FRAME:022874/0914
Effective date: 20061025
|Mar 18, 2013||FPAY||Fee payment|
Year of fee payment: 4